An AI assistant will happily build you a seven-day China itinerary in thirty seconds, complete with a museum visit that closes on Mondays, a temple that shut for renovation two years ago, and a train that does not exist. The itinerary looks finished. It is not. The failure mode has a name, hallucination, and it is not a bug you can prompt your way out of. What you can do is run a short verification routine before you commit money or a day of your trip. Ten minutes of source checking against official pages is usually enough to catch the errors that would otherwise cost you an afternoon standing outside a locked gate.
Why AI invents opening hours so confidently
Language models generate plausible text, not verified text. Opening hours, closing days and ticket rules are exactly the kind of detail that appears in millions of similar sentences, so a model can produce a fluent answer without any grounding in the current page. The output has no signal attached telling you whether it was copied from a 2019 blog post or invented outright. Treat every time, date and price in an AI itinerary as a draft placeholder until you have seen it on a page controlled by the operator.
The ten-minute routine
Work through the itinerary and mark every item that has a time attached: museums, galleries, temples, ticketed attractions, timed-entry slots. Ignore restaurants and streets for now. For each marked item, open a search engine and look for the venue’s own site or its page on a municipal culture and tourism portal. Compare three things only: opening days, last entry, and whether advance booking is required. If the venue has no official page in a language you read, that is itself useful information, and it means you should plan around uncertainty rather than a precise slot.
Do this in one sitting rather than spreading it across the week. The point is to find the two or three items that break your plan, not to audit every line. A single confirmed closure can reshape a whole day, and it is far cheaper to learn that at your desk than at the gate.
What to check on the official page, and what to ignore
On the official page, look for the notice section, which is where temporary closures, maintenance and holiday arrangements appear. Chinese venues often publish adjusted hours around public holidays, and those adjustments are frequently missing from third-party listings. Check the booking channel named by the venue itself rather than a reseller, and confirm whether a passport number is required at purchase. If the venue requires timed entry, book before you build the rest of the day around it.
Ignore aggregator listings with no update date, and ignore anything an AI tells you about a specific number of minutes to allow inside. Duration estimates are judgement, not fact, and you can set your own once the opening window is confirmed.
Anchoring the rest of the itinerary to verified facts
Once the fixed points are confirmed, build transport around them. Rail schedules and ticketing rules change, so check the operator’s own booking channel and the passport rules before you lock in a date. Our guide to buying China train tickets with a foreign passport covers what to have ready. For city days, metro systems are the most predictable part of any plan, and the China metro guide explains how to pay and ride without guessing.
Entry rules deserve the same treatment. Visa-free transit eligibility depends on your nationality, your ports and your onward ticket, and it is not something to take from a chatbot summary. Read the official immigration or embassy page, then cross-check with a written checklist such as the 240-hour visa-free transit checklist so you know which documents to carry.
A prompt pattern that reduces invented detail
Ask the model to separate what it knows from what it is guessing. A useful instruction is to list each attraction with a blank field for opening hours and a note saying the traveller must confirm on the official page. This does not stop hallucination, but it stops the invented hours from looking authoritative in your notes. Then ask for two alternatives per day, so a confirmed closure does not collapse the plan. Finally, ask it to flag any item where the venue is seasonal or has changed location recently, since those are the entries most likely to be wrong.
Build in slack, not precision
Even a verified itinerary meets reality: queues, weather, a long lunch. Leave one unbooked block per day and keep the fixed, timed items to one or two. If a museum slot is the anchor, put nothing fragile after it. If you are moving between cities, allow a buffer around the transfer rather than scheduling an attraction on arrival day. The goal is an itinerary that survives one thing going wrong.
Frequently asked questions
Can I trust AI for a China travel itinerary at all?
Yes, for structure. It is good at grouping sights by district, suggesting a sensible order and estimating how many days a city deserves. It is unreliable for specific times, prices and closure days, so verify those separately.
How do I spot a hallucinated opening time?
Look for suspicious precision, such as an exact closing time with no source, or a claim that a site is open every day. Check the venue’s own page. If the detail appears nowhere except the AI answer, treat it as invented.
What if a venue has no official page I can read?
Assume the hours are uncertain and plan the visit as a flexible block rather than a fixed slot. Ask your accommodation to confirm by phone the day before, and have a nearby alternative ready.
Do I need to re-check everything after I arrive?
Re-check the next day’s timed items each evening. Temporary closures and holiday adjustments are common, and a two-minute confirmation on the venue’s page or at your hotel desk prevents most wasted trips.
Should I use an app or a paper plan?
Either works if the confirmed details live somewhere you can read offline. Before relying on any app, check that it loads without a working local connection, and keep the venue names and addresses saved as text.
Use AI to draft the shape of the trip, then spend ten minutes turning its confident guesses into checked facts. The itinerary you carry should be built from official pages, not from a model’s best impression of them.